Abstract
Oculomotor system controls human eye movements while scanning a scene. One cue used by visual system to make sense of a scene is the border of objects. Our goal is to understand whether oculomotor system has any pre-set predictions while mapping out a scene. To test this hypothesis, we designed a set of 3D scenes by using an integrated 3D projection and eye-tracking system to compare human eye movement patterns during depth cue consistent occlusion (CCO) and inconsistent occlusion (CIO) of a moving target in a pursuit task. In CCO, the moving target gets occluded by an object in front of it or will remain visible if the object is behind. However, in CIO, the moving target gets occluded by an object behind it or will remain visible if the object is in front. It has been known that when a moving target becomes invisible midway on its path, eye pursuit switches to saccades. In this study, we found that least switches to saccades occur in CCO and when the pursuit target is visible. On the other hand, the maximum switches to saccades occur during CIO when the pursuit target is invisible. Saccadic interruption increases slightly in CIO where the target is visible, and even more so in CCO when the target is invisible. We conclude that during scanning a visual scene, human oculomotor system utilizes a set of predictions, perhaps based on accumulated previous experiences, such as "when a moving target goes behind an occluder it should disappear", and when the predictions are challenged, the oculomotor system switches pursuit to saccade. Finally, we present a set of metrics to quantify the interactions between visual-system-based scene segmentation and eye movement patterns, and the interactions between border-ownership and eye movement vectors to establish eye movement correlates of figure-ground segregation and border-ownership.
Meeting abstract presented at VSS 2018